基于卷积核的港口客户细分方法

根据港口客户数据特点,运用信息增益方法对其进行了数据预处理,将其表示为树形结构组织方式,得到216棵客户树;引入卷积核,定义了度量客户树之间相似性的卷积树核;随后,将先前提出的核k-凝聚聚类算法推广到基于卷积核的客户树上,并运用Matlab数据处理工具实现对港口客户数据的聚类分析.分析结果表明,卷积核在港口客户细分中得到了良好的应用效果.

Method of port customer segmentation based on convolution kernels

According to the characteristics of port customer data, a data preprocessing was done applying the method of information gain, and 216 customer trees were got by representing it into tree structure. A convolution tree kernel by measuring the similarity between customer trees was defined by introducing the concept of convolution kernels. Kernel k -aggregate clustering algorithm presented before was extended to customer trees based on convolution kernels and the clustering analysis of port customer data was realized by using Matlab tools. The results of analyses show that application of convolution kernels in port customer segmentation gets good effect.